Causal inference and data fusion in econometrics

Summary Learning about cause and effect is arguably the main goal in applied econometrics. In practice, validity of these causal inferences contingent on a number critical assumptions regarding type data that has been collected, substantive knowledge available phenomenon under investigation. For instance, unobserved confounding factors threaten internal estimates; availability often limited to nonrandom, selection-biased samples; effects need be learned from surrogate experiments with imperfect compliance; extrapolated across structurally heterogeneous populations. A powerful flexible inference framework required order tackle all challenges, which plague essentially any analysis varying degrees. Building structural perspective causality introduced by Haavelmo (1943) graph-theoretic approach proposed Pearl (1995), artificial intelligence (AI) literature developed wide array techniques for allow us leverage information various imperfect, heterogeneous, biased sources (Bareinboim Pearl, 2016). this paper, we review recent advances made have potential contribute econometric methodology along three broad dimensions. First, they provide unified comprehensive learning, above-mentioned problems can addressed generality. Second, due their origin AI, come together sound, efficient, complete (to formally defined) algorithmic criteria automation corresponding identification task. And third, because nonparametric description models approaches build on, combine analytical rigor econometrics flexibility outcomes framework, thus offer valuable complement two streams.